Abstract The three-dimensional hydrodynamics of Galveston Bay were simulated in two periods of several month duration. The physical setting of Galveston Bay is described by synthesis of long-term observations. Several processes in addition to tidal hydrodynamics and baroclinic circulation processes contribute substantially to the observed variability of currents, water level and salinity. The model was therefore forced with realistic water levels, river discharges, winds, coastal buoyancy currents (due to the Mississippi River plume) and surface heat fluxes. Quantitative metrics were used to evaluate model performance against observations and both spatial and temporal variability in tidal and sub-tidal hydrodynamics were generally well represented by the model. Three different unstructured meshes were tested, a triangular mesh that under-resolved the shipping channel, a triangular mesh that resolved it, and a mixed quadrilateral-triangular grid with approximately equivalent resolution. It is shown that salinity and sub-tidal velocity are better predicted when the important topographic features, such as the shipping channel, are resolved. It was necessary to increase the seabed drag roughness in the mixed quadrilateral-triangular grid simulation to attain similar performance to the equivalent triangular mesh.